Fix media resolution and interactive login on the Chrome/CDP path - #1
Open
yingzhouquan3-art wants to merge 6 commits into
Open
Fix media resolution and interactive login on the Chrome/CDP path#1yingzhouquan3-art wants to merge 6 commits into
yingzhouquan3-art wants to merge 6 commits into
Conversation
ffmpeg opens HLS through a proxy when http_proxy/https_proxy are set in
the environment, but the protocol whitelist did not list httpproxy, so it
refused every playlist with:
Protocol 'httpproxy' not on whitelist 'file,http,https,tcp,tls,crypto'
Downloading behind a corporate or local proxy therefore always failed.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The catch-all handler recorded only the exception class name, so a lesson that failed with FileNotFoundError gave no hint about which path was missing. Record the exception message and the failing frame as well, capped so the column stays readable. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
A headless player never issues a real media request without a user gesture, so _media_urls observed no .m3u8/.mp3 traffic and every lesson failed with "No playable audio or HLS request was observed." The Ego browser path already reads getPlayUrl, but the Chrome/CDP path did not. Record the getPlayUrl and get_lookback_list responses through an XHR/fetch hook installed before navigation, and feed the signed URLs they carry through the existing _synthetic_media_events path. Audio-only mp3 is preferred where offered: it needs no HLS assembly. Also skip liveplay.* hosts when sniffing responses. A finished live lesson still advertises its push-stream URL, which 404s, and it would otherwise win over the replay playlist behind get_lookback_list. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
ensure_running() overwrote its own visible argument, so the managed browser was always headless. On platforms with no Ego browser to inherit a session from, that leaves no way to complete a QR-code or SMS login: the window needed to scan the code never appears. Honour the argument instead, and let requested_mode report it so an existing session of the wrong mode is restarted. The default is unchanged - only "auth start --visible" asks for a window, so automation still never steals the user's foreground application. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Audio lessons (/p/course/audio/a_...) never expose a media request the sniffer can see, so they failed with "No playable audio or HLS request was observed." Their source sits in the audio.info.get response, whose data.audio_info.audio_url is already a direct mp3. Record that endpoint alongside getPlayUrl and get_lookback_list — the Ego path's own capture pattern lists it too, so the CDP path was the only one missing it. Also mark PLAY_URL_RECORDER as a raw string: the JS regex it now carries contains \. escapes, which Python flags as invalid escape sequences. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The assertion pinned requested_mode to "headless" for both inputs, which encoded the old always-headless behaviour. With visible honoured as an opt-in, the contract is that it reports what was requested. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
This was referenced Aug 29, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Four fixes found while archiving a purchased course on Windows 11, where
there is no Ego browser to fall back on. Each commit stands alone.
1.
httpproxymissing from the HLS protocol whitelistWith
http_proxy/https_proxyset, every HLS download failed:2. Unexpected download failures lost their cause
The catch-all recorded only the exception class, so a real failure read
Unexpected download failure: FileNotFoundErrorwith no indication ofwhich path was missing. Now the message and failing frame are kept.
3. Media sources were never resolved in the CDP path
A headless player issues no media request without a user gesture, so
_media_urlsobserved nothing and lessons failed with "No playableaudio or HLS request was observed." The Ego path already reads
getPlayUrl; the Chrome/CDP path did not.This records
getPlayUrlandget_lookback_listvia an XHR/fetch hookinstalled before navigation, then feeds the signed URLs through the
existing
_synthetic_media_eventspath. Audio-only mp3 is preferredwhere offered, since it needs no HLS assembly.
It also skips
liveplay.*hosts while sniffing: a finished live lessonstill advertises its push-stream URL, which 404s, and it would otherwise
win over the replay playlist behind
get_lookback_list.Verified on a 50-lesson course: video lessons resolve to mp3 direct
links, live replays to their
playlist_eof.m3u8.4.
auth start --visibleensure_running()overwrote its ownvisibleargument, so the managedbrowser was always headless. Without a window there is no way to finish
a QR-code or SMS login on a platform with no Ego session to inherit.
The default is unchanged — automation still never steals the foreground
application. Only the explicit
auth start --visibleopens a window.Testing
pytest testsgives 63 failed / 102 passed / 5 errors both before andafter these commits, i.e. no new failures. The pre-existing failures look
environment-related (the suite assumes macOS paths); happy to look into
them separately if useful.